Computer Chess Club Archives


Search

Terms

Messages

Subject: Some thoughts re computer analysis...

Author: Francis Monkman

Date: 14:08:06 05/17/99


Analyzing a good number of GM games using a computer, one clearly finds
examples of flawed analysis. On the other hand, one (perhaps as often)
finds examples where it could be said that the computer 'simply doesn't
understand the position', and sees advantage for an already-losing side.
I'm surprised that a strategic principle (ie let your opponent think he's
winning) practised throughout history by human players should prove so
hard (at least at the moment) to conceive against 'the machine'.

A good, simple example of why never to take computer analysis for granted:

Play into any engine the following moves: 1. g3 d5 2. Bg2 c6 3. Nf3 Bg4
4. c4 e6 5. b3 and watch the analysis for a while. I'd be surprised if a
clear advantage to Black isn't registered, probably with ...dxc4 as the
chosen continuation -- most programs would perhaps play this at 40/2?
(Rebel10 no exception, at 200MHz, although the advantage it gives Black
is much less than some programs -- about +0.20. Some register >+1.00.)

Now walk it through the following: 5. ...dxc4 6. bxc4 Bxf3 7. Bxf3 Qd4
8. Qb3 and while the analysis of...Qxa1 initially resembles a Rottweiler
with a sausage, most programs calm down after a little and prefer ...Qd7
or even ...Nd7 (probably best, but still with an excellent game for White).
After 8. ...Qxa1, of course, 9. Qxb7 gives White a clear advantage.

Well, I suppose that's why we give computers opening books -- but then I
guess we all know Deep Blue shouldn't really have won with a 'look-up' sac.

[This is from an interesting game, Ubilava-Gurgenidze Tbilisi,1971 [Gufeld],
where Black played 5. ...Nd7 and after a hard struggle with several winning
chances missed by White, Black lost on time with a winning position!]

Francis




This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.